Context
A unified knowledge graph is worthless if it's hard to get data in and out. Integration is where most asset-data programmes lose budget: writing custom code to ingest an IFC at object level, to keep a Relatics requirements register in sync, to transform a vendor inspection CSV into a typed asset record. That code breaks, rots, and outlives its author. Weaver's model is the opposite: configure pre-built modules, don't write integration code.
Federate through pre-built modules. Don't write integration code.
Explanation
Three integration patterns, in descending order of how much they cover:
Pre-built modules (the default, covers most needs)
Out of the box, Weaver speaks IFC (full object-level ingestion), BCF (issues from any buildingSMART-compatible tool), openCDE (CDE-to-CDE federation), OData (Microsoft enterprise tools), and REST/GraphQL/SPARQL for everything else. Direct, battle-tested connectors exist for Relatics, Autodesk Construction Cloud / Revit, SharePoint, and MinIO/S3 blob storage. Connecting a source is a configuration task: point the module at the source, map a few fields, done. No integration code.
Configurable pipelines (for less common sources)
Less common sources (Maximo, SAP PM, custom ERPs, scheduling tools, requirement-management tools) integrate via configurable pipelines. The pipeline declares how source records map to RDF triples; the engine handles the rest. Pipelines are themselves RDF configuration, readable, versionable, reusable across projects.
Custom connectors (the rare exception)
Genuinely unique sources integrate via custom connectors built against the open API surface. This is the exception, not the rule. Customers and partners regularly contribute connectors back to the platform, so what was once custom often becomes a pre-built module in the next release.
Configuration of integrations is what Certified Weavers do. Adding a new asset class, mapping a new vendor's CSV, hooking up a new authoring tool: hours, by a customer-side operator, never multi-week engineering cycles. The unique value of a unified knowledge graph collapses if customers have to maintain integration code to keep it fed; the architecture is deliberately built so they don't.